Screening Lung Diseases Using Cascaded Feature Generation and Selection Strategies

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• The COVID-19 pandemic necessitates advanced diagnostic tools beyond standard testing.
• Artificial intelligence (AI) offers potential for improved disease identification and monitoring.

Purpose of the Study:

• To develop a smart auxiliary framework using machine learning (ML) for classifying COVID-19, pneumonia, and healthy individuals from X-ray images.
• To enhance the accuracy of ML classifiers through transfer learning (TL) and feature selection techniques.

Main Methods:

• Investigated three TL networks (AlexNet, ResNet101, SqueezeNet) for feature generation.
• Applied feature selection methods (iNCA, iChi2, iMRMR) to refine features.
• Utilized CNN, LDA, and SVM classifiers for image classification.
• Employed stationary wavelet transform and data augmentation to improve model robustness.

Main Results:

• A combination of AlexNet, ResNet101, SqueezeNet, iChi2, and SVM achieved 99.2% classification accuracy.
• AlexNet, ResNet101, SqueezeNet with iChi2 and a proposed CNN achieved 99.0% accuracy.
• Cascaded feature generation and selection strategies significantly impacted classifier performance.

Conclusions:

• The proposed AI framework effectively aids in the identification and monitoring of COVID-19 cases using X-ray images.
• Transfer learning and advanced feature selection significantly improve diagnostic accuracy for respiratory conditions.